Skip to main content

Setting Up Quality Analysis Intelligence (QAI): Best Practices

Written by Moja Bot
Updated over a month ago

Quality Analysis Intelligence (QAI) uses AI to automatically analyze your call recordings, extract key information, and classify calls based on custom criteria you define. This guide will help you set up QAI effectively to get accurate, actionable insights.

Before You Start

Recommended setup order:

  1. Design your dispositions first (Sale, Callback, Spam, etc.)

  2. Create questions for each disposition

  3. Assign priorities

  4. Test and refine

💡 Tip: Start simple with 3-4 core dispositions. You can always add more later.

Step 1: Plan Your Dispositions

Dispositions are the final classification of a call—the most important outcome that happened.

Common dispositions for most businesses:

  • Sale – Customer made a purchase

  • Callback – Customer requested follow-up

  • Not Interested – Customer declined offer

  • Spam/Scam – Invalid or fraudulent call

To add a disposition:

  1. Navigate to your campaign

  2. Scroll to the Call Dispositions section

  3. Click + Add Disposition

  4. Enter:

    • Display name (e.g., "Sale")

    • Priority (1-100+, higher = more important)

    • Match criteria (e.g., "100% of questions must match")

📸 Screenshot needed: Campaign page showing Call Dispositions section

Step 2: Create Effective Questions

Questions help the AI identify which disposition applies to each call.

Golden Rules

✅ Do:

  • Ask specific, single-focus questions

  • Use yes/no questions whenever possible

  • Ask 2-4 questions per disposition for accuracy

  • Be explicit about what you're looking for

❌ Don't:

  • Use compound questions (asking two things at once)

  • Ask open-ended questions without format constraints

  • Rely on one question to determine a disposition

Example: Good vs. Bad Questions

❌ Bad (Compound Question):

"Does the caller own their home and live in the U.S.?"

✅ Good (Separate Questions):

1. "Does the caller own their home?"

2. "Does the caller live in the U.S.?"

Step 3: Build Questions for Key Dispositions

For a "Sale" Disposition

Goal: Confirm a sale happened with high confidence

Recommended questions:

  1. "Did the agent provide a confirmed itinerary or order details?"

  2. "Did the customer provide payment information?"

  3. "Did the customer verbally confirm the purchase?"

Match criteria: Require 100% (all questions = yes)

Priority: 100 (highest)

💡 Why multiple questions? Single questions catch false positives. For example, "Did they discuss pricing?" might match someone just asking about costs, not completing a purchase.

For a "Customer Service" Disposition

Goal: Identify customer service calls vs. sales calls

Recommended questions:

  1. "Did the caller ask to speak to customer service?"

  2. "Was the caller transferred to customer service?"

  3. "Did the caller mention an existing account or order number?"

Match criteria: Require 2 out of 3 questions = yes

Priority: 60 (lower than "Sale")

📸 Screenshot needed: Disposition configuration screen with questions

For a "Spam/Scam" Disposition

Recommended questions:

  1. "Was the call under 120 seconds?"

  2. "Did the caller use profanity or aggressive language?"

  3. "Did the target hang up the call?"

Match criteria: Require 2 out of 3 questions = yes

Priority: 10 (lowest)

Step 4: Understand Priority

When multiple dispositions match a call, Moja selects the one with the highest priority.

Example Scenario

A 5-minute call where:

  1. Customer asks for customer service (matches "Customer Service," Priority 60)

  2. Agent upsells a product

  3. Customer provides payment (matches "Sale," Priority 100)

Result: Call is classified as "Sale" because Priority 100 > 60

Priority Guidelines

Priority Range

Use For

Example

----------------

---------

---------

100+

Sales, Conversions

Sale, Booking, Enrollment

75-99

Qualified Leads

Appointment, Callback

50-74

Service Calls

Customer Service, Billing

25-49

Low Value

Not Interested, Hang-up

1-24

Junk

Spam, Scam, Fraud

💡 Pro Tip: Leave gaps between priorities (100, 80, 60, 40, 20) so you can insert new dispositions later without renumbering.

Step 5: Choose Answer Types

Yes/No Questions

Best for: Binary decisions (Did X happen?)

Setup:

  • Answer type: Yes/No

  • Expected format: yes / no

Example: "Did the customer provide their email address?"

Multiple Choice

Best for: Categorizing responses (Which airline? Which service?)

Setup:

  • Add each option

  • Always include "None" or "Other"

Example: "Which airline did the caller inquire about?"

  • Delta

  • United

  • American

  • None ← Important!

⚠️ Warning: Without a "None" option, the AI is forced to pick an option even when it doesn't apply, creating false data.

Number

Best for: Extracting numeric data (quote amount, age, etc.)

Setup:

  • Answer type: Number

  • Expected format: 0,000.00 (with commas and decimals)

Example: "What was the quoted price?"

Text Answer

Best for: Open-ended extraction (Order ID, email, reason for call)

Setup:

  • Answer type: Text

  • Expected format: Define pattern (e.g., ID-XXXXXXX for 7-digit IDs)

💡 Tip: Avoid text answers for disposition matching. Use them for data extraction, not classification.

Step 6: Test Your Setup

  1. Go to Reports → Filter by your campaign

  2. Add filter: "Disposition is not blank"

  3. Review 5-10 calls

  4. Click into calls to see transcripts and question answers

Check:

  • Are "Sale" calls actually sales?

  • Are "Spam" calls actually spam?

  • Any false positives or missed classifications?

📸 Screenshot needed: Reports page with disposition filter applied

Step 7: Refine Based on Results

If you see false positives:

  • Add more questions to the disposition

  • Make questions more specific

  • Increase match criteria (require 100% instead of 50%)

If dispositions aren't triggering:

  • Check if questions are being answered (view individual calls)

  • Reduce match criteria (2 out of 3 instead of 100%)

  • Rephrase questions more explicitly

If wrong disposition is winning:

  • Adjust priorities (Sale should be highest)

  • Check for duplicate priorities

Common Mistakes to Avoid

❌ Setting All Dispositions to Priority 0

Problem: System can't decide which disposition is most important

Fix: Use a priority hierarchy (100, 80, 60, 40, 20)

❌ Asking Compound Questions

Problem: "Did the customer provide name and email?" – If they only gave one, AI can't answer accurately

Fix: Split into two separate questions

❌ Using Only One Question Per Disposition

Problem: Single questions lead to false positives

Fix: Use 2-4 questions and require 100% match for critical dispositions like "Sale"

❌ Not Including "None" in Multiple Choice

Problem: Forces AI to pick an option even when none apply

Fix: Always add "None," "Other," or "N/A" as an option

Viewing Your Results

Option 1: QAI Analysis Page

Path: Campaign → QAI Analysis → Reports

Best for: High-level question performance analysis

Option 2: Reports Page (Recommended)

Path: Reports → Add Filter → "Disposition is not blank"

Best for: Call-by-call review and exports

📸 Screenshot needed: QAI Analysis page and Reports page views

Option 3: Export for Analysis

Path: Reports → Filter → Export to CSV

Best for: Bulk analysis in Excel or Google Sheets

Need Help?

Next Steps:

Related Articles:

  • Understanding QAI Priorities

  • Creating Custom Questions

  • Exporting QAI Data for Analysis

  • Troubleshooting Common QAI Issues

Last updated: February 24, 2026

Did this answer your question?